Fireside Chat: Copyright, Intellectual Property, and AI Generated Content with Attorney Ruth Carter

Please note the following warning, disclosure, and disclaimer. This interview does not constitute legal advice or create a client or attorney relationship with anyone. The information contained in this interview is presented on an as-is basis with no guarantee of completeness, accuracy, usefulness, timeliness, or of the results obtained from the use of this information, and without warranty of any kind, express or implied, including but not limited to warranties of performance, merchant ability, or fitness for a particular purpose. While we have taken every reasonable precaution to ensure that the content is accurate, errors can occur. In all cases, you should consult with a qualified professional familiar with your particular situation for advice concerning specific matters.

And Ruth Carter, you are that qualified professional. So please tell us who you are and what you do. Hi, Chris. I am Ruth Carter, a licensed attorney based in Arizona. I am an authority on business, intellectual property, and internet law.

And I have recently been bombarded with questions about how AI and the law interact. Have a copyright in the US, you only need two things. You need an original work of authorship that is fixed in a tangible medium. That's it, it doesn't put any qualifications on the quality of the work. Um that tangible medium can be a digital file that requires some type of device or computer to properly perceive it.

Uh, but one thing to note there is no copyright protection for um facts, ideas, or unoriginal uh arrangements of facts. Okay. What's the difference between a copyright uh like a trademark and a patent? Because we hear these terms a lot, is particularly in like AI. Yeah, we hear these terms a lot, and unfortunately, many people who are in the media use these terms uh interchangeably when they're not.

So copyrights apply to original works. So in the time before computers, we would think of things like painting, sculpture, architecture, dance, music. Um, those are things that are covered by copyright. Trademarks apply to what people in marketing would call branding. So that's names of companies, products, logos, slogans, the things you put on the things that you're offering to the public to differentiate what you're offering from your competition.

Um, and it is possible for a trademark to be protected by both trademark and copyright, depending on what you're claiming as your mark. And then patents applies to inventions. So thankfully, most people aren't confusing uh copyright trademarks and patents, but it's the copyright trademark thing where people tend to um misuse or think that the terms are synonymous when they're not. Gotcha. Because you had mentioned there's no copyright protection for ideas and algorithms, but there can be patent protection for those.

Exactly. Okay. So at content marketing world, we were both at Content Marketing World uh not too long ago. You'd said that case law around AI created content is largely settled based on the selfie, right? This selfie here, uh, in the case Naruto versus Slater, uh content created by non-humans is inherently public domain.

Can you outline that case and and what it entailed and and what how it applies to AI? Okay, so here's what happened. Slater is a photographer, he was in Indonesia, and he had put his camera down, and Naruto, this monkey, came along and picked it up and started fiddling with it and ended up taking several pictures, including this selfie. And so this case centers on who owns the copyright in uh this photograph. The photographer who owned the camera says I own it, and um PETA who brought this case on behalf of Naruto says no, the monkey owns it.

And so the ruling in the case was that um the monkey cannot own the copyright in the photo because only humans can own copyrights. So in this situation, because um somebody who is not a human took the photo, it doesn't qualify for copyright protection. And so this photo is in public domain because um it has no copyright, it never had a copyright. So I can show this here, and this is totally fair game. I could I could use this in a blog post even because it is public domain.

Absolutely. You can put it on t-shirts, you can sell it, you can you can make prints, you can do whatever you want with this photo, and nobody can stop you. Um on the flip side, you can't stop anybody from competing with you because you have no rights in it either. Uh one point of clarification. You said a human can hold a copyright.

But can't don't corporations also hold copyrights? Or is that not the case? No. Um copyrights can be held by corporations, by legal entities, but corporation, but you have to have a human involved in the creation of the thing. They have to be primarily involved.

You can use um a computer to you know create your art or create your blog post or whatever is your work, but the computer doesn't own the copyright, even though it was used in the process. Um, yes, a company can own a copyright, but a human has to create the work that is protected by copyright. A human being, not a legal, um, not a you know, not a legal entity is not enough. Got it. Okay.

So why is this a case of you know generation by a non-human outweighing the model of uh software like Microsoft Word? Like, for example, Microsoft Word is is a piece of software, and I work with it to generate content and AI-based tools. I do the same thing to a degree, right? I provide prompts, I provide model tuning. Why would the output of a an AI not be protected in the same way that the output of Microsoft Word is protected by copyright?

So I look at Microsoft Word as a medium. Um, just like we don't give copyright to paint and canvas, we give it to the person who used them. So you're using a tool. Now there are software that will do like auto-completes and things like that. Um, but then in those situations, I would ask, what are you claiming as the work?

If are you creating something that is solely um auto-complete and just whatever it suggests next you're approving? Um in that case, I would ask, is that the entire work or is that like one sentence? Uh, because there's probably not copyright protection in one sentence because that's not enough to qualify as an original work of authorship. Um then I'm thinking about my phone, where like if I'm on Facebook and it's uh suggesting it's giving me like three choices, and I'm still the one deciding which one it's going in. So there's an argument that it's still a human choice as to which options you are going with.

Okay, so that was like the uh the copyright case with um the copyright office on Thaler, where Thaler the the all the copyright office said that um they don't view uh they will not register work produced by a machine or mere mechanical process that operates without any creative input or intervention from a human author. So is there how do how do you how do you define the line between what constitutes because obviously a machine that I there's like I guess a spectrum of generation, there's totally human-made where it's like you and me in Word, like yeah, we're we're we're doing everything pushing the keys. There's totally AI generated where machines just cranking out stuff, and there's you know, you just you don't even give it a prompt, you say just make stuff, um, which is what happens with generative adversarial networks. But then there's this fuzzy ground all in the middle of, well, I did provide a prompt and I did reject, you know, uh stable diffusion, made 25 versions of this this prompt, and I had to be the one to pick which one of the 25 it was because 24 of them sucked. Um where is where is there a line and and how does uh how does someone who's interested in in copywriting their their works, how do they how do you find that line?

So it's important to differentiate in Thayer. What happened there is uh he listed himself on the copyright application as merely the owner of the machine. Um so and he claimed that the machine did all the creating. So from that perspective, um you can't apply for a copyright, or you can't list an author who's not um who's not an owner or um who did not do the actual creating. So that was part of his issue, is that there was no human on the application who created.

Um and then getting into you know, where is the line? If you create something and then run it through a process to get the AI generated result, what you input could be protected by copyright. And then the resulting work that came out of the AI could be a derivative work of your prompt, which still puts the copyright ownership back on the person who created the prompt. So no, that's interesting because obviously for something like if I put in a photo of myself, right? Um that I took, and I say, I want you to put a pink flamingo party hat on me.

Um, actually, let's back up. What is what is a derivative work? I've heard that term a lot, and I would like an official opinion. What is a derivative work? All right.

So to back up even further, when you own a copyright, you get five rights. You get the rights to copy, distribute, display, perform, and to make derivative works. So derivative works are works that are based on an original work. So if you wrote a paragraph in English and put it through Google Translate, that translation would be a derivative work of what you created. The best way I know to explain derivative works is with Star Wars.

So the original work may be the book or the first movie. And whoever owns the copyright gets to decide what derivative works are made. Gotcha. Okay. So in the case of these generative AI models, I put in a photo of myself to say, put a pink flamingo hat on me.

I took the photo. That's pretty clearly a derivative work, right? It's still substantially my work. It'd be like me, you know, scratching off Star Wars on the DVD and calling, you know, Space Wars. That would clearly be a rip-off.

Exactly. How about a text prompt that yields an image? I specify I want a picture of a Korean woman at a desk wearing a gray suit, uh, drinking a cup of coffee in an office at sunrise, highly realistic, 8K resolution. That's my prompt. And the machine says, okay, here's 45 versions.

Please choose one of these 45. And I pick the one that I like. Is that a derivative work? Because it's a totally different medium. So I guess the question would be, is what you put in an original work worthy of copyright, and a list of factors may not be.

So in that situation, I don't see it. Um, based on the information provided, that that resulting photo would be a derivative work. Um, on the flip side, somebody else who has who has taken a photo of a woman in a gray suit drinking coffee at sunrise in an office couldn't say, Hey, you copied me necessarily, um, unless you can prove that the machine actually pulled and copied, because um there are because what you may have only copied in comparison are facts. Like this is what a human looks like, this is what a gray suit looks like, this is what coffee looks like, this is what sunrise looks like. You've only copied the only similarities are factually based, not uh a situation where somebody took this photo and scanned it or Xeroxed it to generate this AI generated image.

Okay. So I guess that means derivative works works in reverse as well. If I take a public domain image and I'm and I modify it, is my modification still in the public domain? You would only qualify for copyright protection in what you added to it. You couldn't claim copyright ownership of the aspects of your work that are from what was public domain.

Okay. So like if I put a uh pink flamingo hat on the Mona Lisa, um I I cannot claim the the derivative work in its entirety as as mine because it's substantially based still on a public domain piece of work. Exactly. Okay. Um I'm gonna get into the the the mathematical formula stuff now as as I'm known to do.

So you mentioned um autocomplete, right? So autocomplete is based on a uh a model called LSTMs, long short-term memory, and essentially they memorize the the unique patterns of words that you put in onto your phone over time and essentially spit that back and say the probability of you typing lawsuit immediately after the word copyright is pretty high. So we're gonna auto-suggest that as opposed to uh copyright you know layer cake. Uh that it's probably statistically less likely. If I'm using autocomplete on a phone and I'm writing a paragraph, a poem or something using autocomplete, how where does that sort of fall in the in the spectrum of copyright?

Because a machine, I am I am making judgments about which word to suggest, but I am not the ones actually writing at that point. The machine and I are writing together. So the machine learned based on what you historically put in. So the argument could be that because it is based on past inputs from you, that you're still the writer, and you're still the one who is accepting the auto-complete suggestions. Okay.

You're the one who save or send or whatever to create that work. Okay. So that that puts an interesting twist then on AI models. So there are these really large models like stable diffusion or GPT 3, these you know, these massive models that you get out of the box from a company like OpenAI or Hugging Face. Um that that is its name.

Great name. It's a great name. Um, and then there's a process in model management called fine-tuning, where you say, okay, I've got this large corpus. Uh, well, you know, say it's the the GPT Neo X20B model from Hugging Face. And then I say, I'm gonna take my blog, which is only like 2,000 posts, like it's a very small compared to the 800 million pieces of text that was trained on.

But I can say I want this model to be tuned to most closely mimic my blog. I'm gonna I want it to reweight all of it the its likely outcomes based on my copyrighted works. So I'm taking a public domain model, but I'm tuning it. In that instance, again, how does how does this look from the eyes of the law? Because now it's no longer the stock off the shelf model.

Now it is it is almost flavored with my copyrighted works. Right. So I wouldn't be surprised if we have future lawsuits related to copyright and AI to parse out some of these things. But my best guess is the law may look at these uh AI softwares similar to Microsoft Word, similar to paint and canvas, and this is a tool, but it still comes down to how the human is using it that produces the results. Okay.

So if I get from one of these many, many, many, many, many tools that are available, um, an AI generated blog post. Clearly, if I just copy and paste that on my blog, because it's purely machine generated, I put in a keyword, right? Um, and and a keyword is definitely not enough to qualify as authorship, like you know, uh intellectual property law, and it would spit out a blog post. I copy and paste that in its entirety. That that's almost certainly public domain.

Yes, that would be my best guess. How uh and I know there's no mathematical number here, how much of it would I have to rewrite as a person to get copyright to apply to it again? All right. So let me start by just speaking as a human, and that to date, most AI generated blog posts are crap. I remember when AI tried to like make up pickup lines and it was hilarious and disturbing.

So start there. So from that perspective, you would have to change it a lot to be something worth using. Um, but to be copyrightable, we're getting into where's the mathematical equation, doesn't exist in law, they don't like using that. So they would look at um where is your original thought? Because there is no copyright in facts.

So if this whole blog post just generated facts, like let's say you said, give me a blog post about dogs, and it said, well, dogs have four paws and ears and are can be pets. Well, okay, all you know, it need if the only things you create copied are just the facts, well, there's no all you did was you know, it's like using an encyclopedia. So you probably wouldn't get copyright protection in you know, in what you create unless until you add in original work of authorship, your original thoughts, and then that would be protected by copyright. Gotcha. Okay.

So in because I remember uh I think I mentioned there's a case called Torasoft versus Drosnin, and one of the things that they had said in that case was um the owner of the software and not the user held the copyright if the machine if the software is doing the the machine the lion's share of the work. Um they didn't really explain what the lion what a lion's share is. No, they didn't. And in in that case, and the law's really good at that. Is they they leave the door open for future litigation.

Um I took a look at that case, and what they said was because the issue was are the could the defendant use printouts from this software without violating copyright. So um the software owner was saying, well, those are derivative works because it came through my software, therefore I own it. And so the in that specific case, the law said that the uh software was only doing an unoriginal arrangement of facts, and so there was no copyright in the printouts, and so there was no violation of copyright by using said printouts. There may have been copyright in that software code. Um even binary code, zeros and ones can be sufficient to be protected by copyright, but the um but in this specific case, it sounds like it didn't it didn't get there.

So um there was no copyright violation in the defendant using the printouts in his book. Got it. Okay. Um one of the things that I think is challenging is that the quality of what machines are starting to create is is getting away from absolute crap uh and getting into mediocre to decent, right? So this is an example.

I just put this in. I put in my search terms, implications of AI on copyright law and what this one particular series, this is Go Charlie, spit out, is readable and coherent. It is factually incorrect. So there are, you know, certainly if I was a if I was a human who wanted to use it, I was like, okay, I clearly need to do some editing and cleanup. But from a usability perspective, it's not bad, right?

So in terms of how much I would have to modify, I would probably have to modify the factual stuff, obviously. Obviously. But in terms of tone and the way it's generating content, it's it's not bad. So hence the my question about like how do we determine when something re-qualifies for copyright? Because this is okay-ish.

Except for the fact that it's you know false. Yes. Um but I think if you wanted to use something like Go Charlie to put in an idea to see what it pumps out for ideas or potential like subheadings, I I don't necessarily that's not a bad way to go in terms of looking for ideas to cover the scope of a topic. Um as we previously exchanged emails, um there if you create a blog post based on an outline, you still would probably have copyright in your resulting work uh because you added your original thoughts to it. So I I don't know if we're quite you know, I don't know if this would get us there in terms of having something worth taking back to the court to say let's reexamine this AI copyright machine involvement uh situation but I think it's I think it's uh a way to look at these AI tools as ways to assist content creators in terms of what they create um I know that doesn't really address your copyright question but that's the perspective that that first came to my mind when I looked at what you put on the screen.

Got it if I'm a company or person and I'm using these tools to generate uh AI based content. For example, there's a there was an AI painting that actually won uh an art contest award and the art contest is now like okay we're we're prohibiting um AI generated content and a bunch of services like Unsplash and and Pixabay and stuff are all saying yeah we're not accepting you know submissions of AI generated content either which to me that at least says okay from a quality perspective it's it's there it's enough that you have to make that rule now is as you know previously like that's not gonna win anything. So you don't need to worry about machines entering. If I am creating AI based stuff, how could I protect uh my company's AI generated content? Obviously um if it's purely machine generated it's not uh cut over to under traditional copyright law what about what are the my other options okay couple of thoughts one I want to go back and look at that art contests rules because all the ones that I've looked at to date, not including this one um said like you can only enter in your own art.

So I have my first question is did this art piece that won did it qualify to be entered? Again, I'm not their lawyer, I'm not involved. I didn't write those rules. Um, but that's what I'm curious about. Back to your real question about how does a company protect its AI generated content.

So as we just discussed, it doesn't look like we're gonna have full pieces of work that are fully AI generated that are usable yet. Um it would be hard to say that at least in terms of like a blog post, like a company isn't gonna be doing that. Um, but if you're if but if you are putting out something that is a hundred percent AI generated, let's say like on Instagram, which is very image focused, and a lot of people are creating images, you probably can't. Um you can create you can protect the content, the the verbiage that you put with it if it's if it qualifies as an original work. Um but the image itself, probably not.

There would be nothing you could do to stop somebody from using something that you don't own any rights in. So unless you can prove that it unless it's unless you've somehow prove that this is an original prompt that you put in the original prompt that is copyright protected and the results of the AI gen, you know, the AI generation um is a derivative work, maybe. Um that might be your only your only way to claim protection is to say that's a derivative work of something that is copyright protected. Otherwise, you have to accept that potentially as a risk of using AI to create content. If I slap a big uh watermark on it, then at least my trademark protects the piece, though, right?

If you put your trademark on it and somebody uses it with your trademark, then that's a trademark issue. That's not a copyright issue. But yes, you could do that. But at that point, I could engage your services to sue them. Potentially, yes.

Um, or if you use AI to generate a trademark and you are using it as such, you would have trademark rights. But if somebody takes that trademark and starts using it as their trademark on a completely different type of product or service, that may be okay. Um it is possible for two companies completely separate to have the same trademark um without there being a violation. That's why we can have Delta Dental, Delta Airlines, Delta Faucet. No one thinks that that's that's one company.

We know it's three separate companies. That's true. Yeah, I mean it'd be very strange if it was the same company. Like here, the spite's gonna be miserable anyway, so let's do some gender work qualifying. Exactly.

Okay, so that's interesting then because so if if companies that are if we know companies are using AI generated content, if I were to you know through my own tools, as Google is doing, identify that this is machine generated content. Um, could I reuse it without their permission? I'll give you an example. The the um this this is another copy uh the AI service, and you have all these companies' logos on here. They're clearly saying, you know, these companies use our software.

Well, clearly you're now declaring um these companies are using are creating at least some percentage of their content as machine generated. Okay, but all that tells me is that they're using the tool that's not telling me how they're using the tool. Okay, got it. Is there a legal risk as a company if you publicly state that you use AI uh in your content generation? Could you know could someone else challenge that company and say, yeah, your content should be in the public domain?

I guess I would wonder how much money do you have? And do you really want to do that? Um, because I mean, you know lawyers and lawsuits are not cheap. So I guess my question would be, what makes you think that well, what works are you claiming our public domain, and what's your evidence for that? Right.

And do you really want to challenge it? Like what I mean, if it was my client who came to me with wanting to do this, I would wonder like, are you wearing a tinfoil hat? Um I question would be like, what's the upside in challenging this? Um on the flip side of that, then if I'm the company that is using these tools in my process, how should I think about documenting um that I used humans, um, at least in a part of the process, so that if I ever am challenged by some tinfoil hat person with a lot of money, you know, like some you know, random billionaire on Twitter, um, what evidence would I need to provide to debase that claim? I would look at like what is your company's process for creating content, um, where you probably have things like drafts and you have people who are paid to create content, and you can look at their files and you can look at um we don't really keep minutes at meetings anymore, but if there were minutes from meetings, um things like that, uh, just to see are there discussions about how do we use AI and how do we transform it?

Um one of the things we talked about earlier had to do with Disney. And I could totally see Disney using, I was thinking for some reason I started thinking about Lion King. Um, you know, could look at like, okay, what does a lion look like? How does a lion move? And so to make it authentic to make the um the animation accurately reflect how does this animal move in reality, and that would be one way to assist their creation um without putting themselves in a situation where you know, and a claim of, oh, you know, an AI created this entire movie, which I I don't think we have AI that that's good enough yet to do anything like that.

Not yet. We're getting there. We're uh this morning Google debuted Image NVideo, which is uh take a prompt and it'll create you know a 10-second snippet of video out of it. And so the that we're we're gonna be there at some point. All right.

Um if I see a competitor's content, like I see a competitor's ad, for example, I'm maybe in spy foo the the SEO software, and I can pull copies of you know competitors uh top performing images, and I look at that and I completely generate something new that's materially similar to their ad, like an image of a woman working at a desk, reading a newspaper, sipping a coffee cup at in an office wearing a gray suit in the morning. You were saying earlier this this is still they still would not be able to pursue a copyright claim on their ad because this is not their ad image. This is me interpreting their ad image as a prompt for a machine. Potentially, they wouldn't have to probably potentially bring a lawsuit, put both pictures side by side and say, make the argument that you used AI like a Xerox machine. Um is that likely to happen?

Um, whereas let's say another situation, let's say you were a photographer and you saw this really cool maternity photo where somebody was like posing on a mountain and in with wildflowers around them, and they were holding their belly in a certain way. Um, and you would say you took a pregnant person up to a mountaintop with wildflowers and you know, posed them in the same way, and you generated two very similar pictures. The person who gave you the inspiration cannot sue you for copyright infringement because the only thing you copied was the idea of this is the photo. Uh, this is this is the pose, this is the setting. And so it would tackle the issue of that gray area between what is the difference between copying an idea versus copying a work.

Got it. Okay. Suppose a machine generated image that, you know, maybe somebody put in um, you know, Korean man sitting at desk with microphone, uh, you know, with bookcase behind, uh recording a podcast, you know, generated an image that, say, to a facial recognition algorithm would flag it as me. Do I have any recourse to protect my likeness? Um, even though that isn't me, it was generated, but it it it's passed is so strongly resembles me that it facial recognition would say it was me.

Right. It's a challenging question. Um, because it sounds like they didn't strive to copy you. It just was a happy accident. Uh it's like, you know, and so they could show like, hey, we didn't copy Chris, the fact that it came out similar to Chris is just you know serendipity.

Um it made me think of my friend, uh my friend David Rower looks remarkably similar to Robert Downey Jr. And there was a picture of him like standing next to a poster of Robert Downey Jr. Um like in a movie theater, and Facebook suggested that Robert Downey Jr.'s face was his. Um so you know, it happens. There are people who I think your connection, either your connection or my connection just dropped.

One of the two. There you are, now you're back. One of the two. Probably mine. Okay.

Naturally occurring twinsies, you know, they look similar, but they're not related. Happens. Right. Okay. There's no and nothing legally wrong happened.

Got it. Okay. Given all these issues around copyright and machine generator, do you think there will be a premium on uh human-produced content that has a doc has documentation says a human we certify? It's like just certified organic. Um we certify that this content is organic and human generated, only human generated, no machines were involved.

Do you think if so, is that something you think like your creators should be thinking about as a differentiator? Oh my god, certified organic content. I love that. You you have to do something with that phrase. Um the one challenge with that is no machines used at all means we couldn't use computers.

So how would it get no generative processes? How about that? Okay, no generative processes. Um we may have to have like a little ad, you know. I can't stop being a lawyer.

I'm like a little star, like you know, publishing on the internet is not generative. Um so um get on that. You know, I don't it that's a hard question because on one hand, you don't have to deal with the legal AI issues. On the flip side, if you're just using it as a tool, um it's probably there's a risk that the content it's gonna them. You might have similarly um valued content, but one might cost a lot more to create because it's gonna take the solely human producer longer versus then letting an AI tool help.

So maybe. Um, I think it may come down more to the value of the brand or the value of the person who is creating that's going to be um adding the value, not merely that fully a human with their fingers created this content. Gotcha. Although I mean we do know that there certainly is a market for uh you know goods and services that are artisanal handcrafted, you know, organic, gluten-free, et cetera. Um artisanal content, handcrafted content.

These are all terms you have to find a home for, Chris. Okay. Um, so we we've covered a lot of AI. It sounds like though, that if you want to uh err on the safe side as a as a company creating content, you should have humans substantially involved in in the creation. AI can help, but AI should not be in the driver's seat, is what I would sum up what we've talked about.

Exactly. Okay. And if you've got purely AI generated content, we have to acknowledge there is no copyright on it. Exactly. There is no copyright in solely AI generated content.

Okay. If people are I've heard this and are now like, oh crap, I need a lawyer. Um, how do they get in touch with you? Oh, yes. One of the upsides of copyright law is it's federal, so I can actually help anybody anywhere in the US who wants to talk about their copyright issues.

So I love that. Um, so the best place to find me is geeklawfirm.com. We'll get you connected to to me, my content, my socials, all those things. Awesome. Thank you so much for all of your information, not advice, information today.

Today it has been a pleasure having you on. Uh, likewise, thank you so much. And let's go ahead and hit our